OpenAI Codex Cloud: Tasks That Run With Laptop Closed
At DevDay OpenAI made Codex cloud environments persistent and shareable, with tasks you can follow from a phone. Here is what changed and what to check before moving a repo.
OpenAI Codex cloud environments are now persistent and reusable across the desktop, browser and mobile apps, which means a coding task you start can keep running after you close your laptop. OpenAI announced the change at DevDay on 29 September, and The Decoder reports it is available on Plus, Pro, Business, Enterprise, Edu and Healthcare plans.
What changed in OpenAI Codex cloud
The headline is that the environment becomes an object you keep. According to the Runtime Wire keynote roundup, reusable environments hold the repository, its dependencies and the access settings, so the next task does not start from a blank machine. The Decoder adds that a team can share one configuration with approved settings and permissions.
The rest of the Codex announcements, as listed in the OpenAI developer community thread:
Cloud tasks that execute with the laptop closed, reachable from other devices.
A Codex CLI update with two-way voice and an
/agentsview for tracking delegated work.A code review experience that gives summaries and diffs and lets you ask questions about a change, plus automatic cloud-based reviews. Runtime Wire says GitHub support is generally available and GitLab is in preview.
What it means if you build with coding agents
The change is about where the work runs. A local agent dies with your laptop lid. A cloud agent keeps going, and somebody else's machine now holds a copy of your code, your dependency setup and whatever credentials the environment needs. Our piece on cloud versus local coding agents walks through that trade in general terms.
Before you point a real repository at a cloud environment, answer four questions:
Which secrets does the environment hold, and could a task read them out?
What network access does it have?
Who approves a change before it reaches your main branch?
How will you review output you did not watch being produced? The checklist in how to review AI-generated code before you ship it applies unchanged.
Long unattended tasks also spend usage while nobody is looking. If you run coding agents on a budget, set the limit first, as described in how to set a budget cap on an AI coding agent.
A low-risk first run
If you want to try it, treat the first task as a test of the setup, not of the model.
Pick a repository with nothing sensitive in it, or a throwaway fork.
Give the environment no production credentials. If the task needs an API key, create a scoped one that can be revoked in a minute.
Choose a small, checkable job such as fixing lint errors, adding tests for one function or updating the docs.
Close the laptop, come back, and read the diff the way you would read a stranger's pull request.
Note how much usage the task consumed so you can judge what a bigger job would cost.
The new /agents view in the CLI is the part worth watching if you delegate several jobs at once. It exists to answer "what is still running and what finished", the question that gets hard once more than one agent is working.
What the sources do not say
None of the coverage I read this run gave pricing or usage limits specific to cloud tasks, or a cap on how long a task can run. OpenAI's own recap page returned an access error to my fetch tool, so everything above rests on trade coverage and the developer community thread, not on the original announcement. Check the plan page for your tier before relying on any of it.
For how Codex stacks up against other tools, see Claude Code vs Cursor vs Codex, and for the wider landscape start at our AI coding tools guide.
How did this land?
About the author
Senior Editor, AI & Product
Cecilia leads the Swarmz editorial desk. She has spent a decade turning complex AI and product topics into writing people actually finish, and she owns the blog's quality bar.
